On Three AWE Applications: Grammatical and Mechanical Error Detection, Abstract Writing, and Cause-and-Effect Writing

Automated writing evaluation (AWE) systems have long been catering to the needs of writing instructors to provide timely feedback to students’ writing. While popular commercialized AWE systems (e.g., Criterion, MyAccess!,... [ view full abstract ]

Automated writing evaluation (AWE) systems have long been catering to the needs of writing instructors to provide timely feedback to students’ writing. While popular commercialized AWE systems (e.g., Criterion, MyAccess!, Grammarly) mostly, and effectively, address grammatical and mechanical errors, to our knowledge, only a few AWE systems are capable of providing feedback on genre-specific writing. In this study, we present three AWE applications along with their performance evaluations as informed by design-based research (DBR) principles for their continuous iterative refinement. CyWrite detects grammatical and mechanical errors for ESL/EFL writers. A module in CyWrite provides feedback on the use of lexical bundles and verb categories in rhetorical moves in engineering journal article abstract writing.  Another module evaluates written cause-effect explanations. We assessed the performance of these AWE applications using three corpora: two corpora of essays written by two groups of undergraduate English learners and a corpus of engineering journal article abstracts. The performance-assessing measures were precision, recall, F-score, and accuracy. Our findings show that the three applications perform well compared to one AWE tool taken as a standard; however, their performances still need to be improved for better language detections and learning outcomes. 

This presentation introduces three automated writing evaluation systems and evaluates their performances following design-based research principles. We explore CyWrite’s capacity in detecting (a) grammatical and mechanical... [ view full abstract ]

This presentation introduces three automated writing evaluation systems and evaluates their performances following design-based research principles. We explore CyWrite’s capacity in detecting (a) grammatical and mechanical errors, (b) lexical bundles and verb categories in rhetorical moves in engineering journal article abstract writing, and (c) written cause-and-effect relationships and expressions.